The data center industry keeps talking about power as if the issue is simply supply. Utilities are under pressure, grid infrastructure is strained and new AI workloads are creating levels of demand that many markets were not built to absorb this quickly. Those issues are real, but the more immediate constraint is timing.
Data centers and power infrastructure are being built on completely different schedules, and the gap between those schedules is becoming one of the biggest constraints on AI growth. A data center can move from groundbreaking to operations in a few years, while the power infrastructure needed to support that same facility often takes much longer. Interconnection studies, transmission upgrades, substation work, permitting and utility queue delays can push timelines out five to ten years, sometimes more.
That gap is no longer just a planning inconvenience. It is shaping where AI infrastructure can actually get built, which projects can move forward and which sites remain theoretical long after the real estate has been secured. The industry’s default answer of building more generation only gets part of the way there because new supply does not solve the bottleneck if the power cannot reach the facility when the facility is ready to use it.
New generation projects also face many of the same interconnection and transmission constraints that are already slowing down data center development. Recent capacity market pressure should be read less as a sign that the country has simply run out of power and more as a sign that the system was not designed for this kind of demand to arrive this quickly. AI has changed the speed of infrastructure demand, but grid infrastructure cannot always move at the same pace as capital, customers or compute.
The result is that site selection has changed. It is no longer just a question of land, tax incentives, fiber access or construction cost. Those factors still matter, but power timing now sits at the center of the decision. A site with a longer path to interconnection may look attractive on paper and still be years away from supporting real AI workloads, while a less obvious site with existing power access, nearby substation capacity or a cleared interconnection path may be more valuable because it shortens the wait.
This is one reason markets like Texas have become so important to the data center buildout. The appeal is not simply that power exists there. It is that certain projects can move through the power timeline faster than they can in other regions. For an AI company trying to deploy capacity, that difference can determine whether a facility is useful in the next business cycle or stuck in a queue while demand moves elsewhere.
The same logic is changing how companies think about new construction versus colocation. Building from the ground up gives an operator more control, but it also means taking on the full power timeline. Colocating in a facility that has already secured capacity can remove years of uncertainty, assuming the facility can support the density, cooling and resilience AI workloads require. For many deployments, inherited power access is becoming as important as the building itself.
This is especially true as rack densities climb. AI infrastructure is not asking for the same power profile as traditional enterprise computing, and a facility that was sufficient for yesterday’s workloads may not be ready for racks pulling 50 to 100 kilowatts. That makes timing even more important because the question is not only whether a site has power, but whether it has the right kind of power available when the customer needs it.
The industry will add generation, and utilities are already investing at a scale that would have seemed unrealistic even a few years ago. But generation is only part of the equation. The harder work is getting that power through the grid, through the interconnection process and into facilities built for the next generation of computing.
Until that timeline compresses, the limiting factor for AI infrastructure will not only be how much power exists. It will be how long it takes to get usable power to the places where AI capacity is actually being built.
